Reimbursement for AI Software in Radiology: Challenges and Solutions

With over 200 AI applications already cleared by the FDA, the potential use cases for AI in radiology are vast and diverse. While image interpretation AI applications have garnered the most attention, there is a growing recognition of the significant role that non-interpretative AI applications can play in improving efficiency, safety, and overall quality of radiological services:

• Image interpretation AI applications enhance image processing and interpretation, improving accuracy and sensitivity in detecting radiological abnormalities.
• These applications automate quantification and characterization, reducing variability between radiologists and enabling more precise monitoring of disease progression.
• AI algorithms, particularly deep learning models, can denoise 3D magnetic resonance images, leading to faster scanning times without compromising image quality.
• Improved protocoling, worklist prioritization, and reduced radiation exposure are additional benefits of AI applications in radiology.

Enhancing Radiological Diagnosis with AI Applications

Despite the promise AI applications hold for radiology, their widespread adoption faces several barriers. Integrating AI applications into existing radiology workflows can be complex and time-consuming. Radiologists and other users of AI software may be resistant to embracing new working methods required by these applications. Additionally, healthcare institutions may view investment in radiology AI applications as irrecoverable overhead costs that do not directly lead to cost savings or generate additional revenue.

A critical question that arises is who will bear the financial burden of adopting AI applications in radiology. In most mature healthcare systems, healthcare providers are paid through a prospective payment system. However, these payment systems are often slow to account for the increased costs associated with adopting innovative technologies, potentially hindering the adoption of AI in radiology. On the other hand, payers may be concerned about the overuse of AI technologies if they are paid for separately. Striking the right balance between paying for AI applications separately from the underlying imaging study and ensuring appropriate reimbursement is a challenging task for healthcare systems.

Barriers to Adoption and Financial Burden

While the payment pathways for radiology AI applications are still in their early stages of development, some progress has been made. In the United Kingdom, certain radiology AI applications are already being paid for separately, particularly through the National Health Service (NHS), which reimburses radiology software applications provided as Software-as-a-Service (SaaS).

Existing payment pathways for radiology AI applications in the United States are determined by:

1. Centers for Medicare and Medicaid Services (CMS): is responsible for determining the payment pathways for radiology AI applications in the United States.
2. Inpatient Prospective Payment System (IPPS): The IPPS governs the coverage and payment for healthcare services provided in the inpatient setting. It requires appropriate coding, coverage policies, and payment mechanisms to ensure successful reimbursement.
3. Hospital Outpatient Prospective Payment System (OPPS): The OPPS governs the coverage and payment for healthcare services provided in the hospital outpatient setting. It also requires appropriate coding, coverage policies, and payment mechanisms for successful reimbursement.
4. Medicare Physician Fee Schedule (MPFS): The MPFS governs the coverage and payment for healthcare services provided by physicians. It establishes payment amounts based on Relative Value Units (RVUs) assigned to specific procedures.

Coding, the use of alphanumerical codes to describe services or technologies, is a prerequisite for billing payers. CMS establishes coverage policies that define when a service or technology is eligible for payment, usually based on the criteria of being "reasonable and necessary." The actual payment amounts are determined based on various factors, such as the costs incurred by providers in delivering the service.

Reimbursement Challenges and Payment Pathways

The OPPS and MPFS utilize Current Procedural Terminology (CPT) codes, while the IPPS employs International Classification of Disease (ICD) codes. In the inpatient setting, services are grouped into Medicare Severity Diagnosis Related Groups (MS-DRGs) for payment determination. New technologies meeting certain criteria can receive temporary additional payments called New Technology Add-on Payments (NTAP). The OPPS employs Ambulatory Payment Classification (APC) grouping, and transitional pass-through payments are available for new single-use physical medical devices. The MPFS sets payments for each CPT based on Relative Value Units (RVUs).

Outside of the United States, countries like the United Kingdom and Germany have their own mechanisms for paying for radiology AI applications. The MedTech Funding Mandate (MTFM) in the UK operates as a special policy mechanism to accelerate access to innovative medical devices and digital products. In Germany, new technologies used in different healthcare settings secure temporary add-on payments or listings in the Uniform Assessment Standard (EBM) catalog after a positive health technology assessment. Japan considers software applications approved by the Pharmaceutical and Medical Devices Agency (PMDA) as medical devices eligible for reimbursement.

In the past five years, CMS (Centers for Medicare and Medicaid Services) has evaluated eight radiology software applications that have been FDA-approved. Out of these, six applications received positive coverage and payment decisions from CMS, while two were rejected. The coverage and payment decisions varied depending on the setting in which the applications were used.

CMS Evaluation and Coverage Decisions

In the inpatient setting, technologies need to meet the "substantial clinical improvement" criterion to qualify for separate payments. Among the evaluated applications, only ContaCT (Viz LVO) received NTAP (New Technology Add-on Payment) status. ContaCT is used for detecting large vessel occlusion (LVO) on CT angiogram images. Evidence presented for ContaCT demonstrated its ability to reduce the time-to-diagnose LVO, leading to improved clinical outcomes for patients. However, Rapid ASPECTS and Aidoc Briefcase, which aimed to characterize brain tissue in stroke patients and triage suspected pulmonary embolism cases, respectively, were both rejected. While these applications showed improved time-to-diagnosis, CMS determined that they did not meet the substantial clinical improvement criterion or provide sufficient evidence of superior clinical outcomes compared to existing standards of care.

The success of applications for securing separate payment in the hospital outpatient setting can be attributed to their ability to provide new diagnostic information or improve the diagnostic performance of the underlying imaging study. These applications offer implicit value by enabling non-invasive diagnostic procedures, avoiding additional invasive diagnostic tests, or providing superior diagnostic performance compared to existing models.

Proposed Framework for Reimbursement Criteria

In summary, the reimbursement pathways for radiology AI applications in the United States vary depending on the setting of use. In the inpatient setting, applications need to meet the substantial clinical improvement criterion, while in the hospital outpatient setting, separate payments can be secured without meeting this criterion. The decisions made by CMS highlight the importance of providing robust evidence of clinical improvement and superior outcomes when seeking reimbursement for radiology AI applications.

The rapid advancement of artificial intelligence in radiology has raised an important question: who should bear the financial burden of implementing technologies? Payers face the challenge of balancing the need to fund innovative, cost-increasing technologies against the potential impact on their budgets. This article explores the complexities surrounding reimbursement for AI in radiology and proposes a framework to guide payers in making informed decisions.

Prospective payment systems, which determine payments based on historical cost data, often fail to account for the additional costs associated with innovative technologies. This creates a dilemma for providers: they either bear the economic burden, incurring losses for each episode of care involving the technology or choose not to adopt it due to the potential negative financial impact. Neither scenario is conducive to the successful integration of innovative technologies. Research shows that separate payment for innovative technologies positively influences their adoption.

The number of approved radiology AI applications has seen an unprecedented surge, with 210 applications approved as of November 2022, compared to just six at the end of 2017. These applications offer diverse benefits, ranging from improving consistency in reporting to providing diagnostic information beyond human capabilities. The beneficiaries of these applications can vary as well, including radiologists, patients, and society as a whole. As a result, not all AI applications merit separate payment, and a careful evaluation is necessary to determine their value.

While paying separately for radiology software applications seems logical, there is a concern that providers may overuse technology for economic reasons. An example of this is the exponential growth of computer-aided detection (CAD) in mammography, which led to substantial additional expenditures without improving diagnostic performance. To avoid such pitfalls, clear criteria must be established to determine which radiology AI applications should be separately payable and which ones should be bundled within the underlying imaging study.

Currently, payers lack specific criteria for radiology AI applications and often rely on existing reimbursement pathways, which may not adequately address the nuances of cutting-edge AI technologies. The Republic of Korea stands out as the only advanced healthcare system that has considered developing reimbursement criteria specific to radiology AI applications. Their proposed criteria suggest separate reimbursement for AI applications that provide new diagnostic information, backed by evidence of clinical or cost-effectiveness.

To address the need for clear reimbursement criteria, a proposed framework is presented. The framework considers two key aspects: the nature of the benefits offered by the application and the availability of evidence supporting its clinical and/or economic outcomes. Applications that primarily improve provider efficiency or convenience may not require additional payment. However, applications that enhance diagnostic performance or provide new diagnostic information merit separate reimbursement, especially when supported by evidence of clinical and/or economic benefits.

Conclusion

Reimbursement coverage policies significantly impact the adoption of radiology AI applications, making it crucial for payers to develop clear and consistent criteria. Collaboration between payers and technology developers is essential to establish evidence requirements based on target product profiles. Although the frameworks and evidence standards for pharmaceuticals cannot be directly transposed to radiology AI applications, valuable lessons can be learned from past experiences. Payers must embrace high-quality real-world studies to generate robust evidence for diagnostics, recognizing their unique iterative nature.

AI Software for Radiology

At Rayscape, we develop a software platform that is using Artificial Intelligence to help radiologists analyze medical images faster and more efficiently. Our AI-based algorithms are seamlessly integrated into the doctor’s workflow, helping them analyze chest X-rays and lung CTs more efficiently. Our team is dedicated to creating the best digital radiology system for medical imaging interpretation. Rayscape AI includes pathologies detection, localization, and automatic measurements, all of them sent directly to the hospital PACS system.

Rayscape's in-house created algorithms cover all of a clinician’s most common requirements when it comes to chest image analysis and pulmonary nodules evaluation. Our solution is built around the radiologist’s everyday workflow to increase productivity and efficiency.

References:

• American College of Radiology. ACR Data Science Institute AI Central. (2022).
• Makary MS, Vitellas CA. Artificial Intelligence in Radiology: Current Applications and Future Technologies. Health Manage. 2021;21:205–208.
• Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts HJ. Artificial intelligence in radiology. Nat. Rev. Cancer. 2018;18:500–510. doi: 10.1038/s41568-018-0016-5.
• Jiang D, et al. Denoising of 3D magnetic resonance images with multi-channel residual learning of convolutional neural network. Jpn J. Radio. 2018;36:566–574. doi: 10.1007/s11604-018-0758-8.
• Kwee TC, Kwee RM. Workload of diagnostic radiologists in the foreseeable future based on recent scientific advances: growth expectations and role of artificial intelligence. Insights Imaging. 2021;12:88. doi: 10.1186/s13244-021-01031-4.
• Tadavarthi Y, et al. The state of radiology AI: Considerations for purchase decisions and current market offerings. Radio. Artif. Intell. 2020;2:e200004. doi: 10.1148/ryai.2020200004.
• Adler-Milstein, J. et al. Meeting the Moment: Addressing Barriers and Facilitating Clinical Adoption of Artificial Intelligence in Medical Diagnosis.
• Centre for Medicare and Medicaid Services. Hospital Outpatient Prospective Payment and Ambulatory Surgical Center Payment Systems and Quality Reporting Programs - Proposed rule 2023. 87, 71748–72310 (2022).
• Parikh RB, Helmchen LA. Paying for artificial intelligence in medicine. NPJ Digit. Med. 2022;5:1–5. doi: 10.1038/s41746-022-00609-6.
• Kuo, T. & Manaker, S. Reimbursement Strategies and CPT Codes for Device. In Academic Entrepreneurship for Medical and Health Scientists1, 1–10 (2019).
• Chen MM, Golding LP, Nicola GN. Who will pay for AI? Radio. Artif. Intell. 2021;3:e210030. doi: 10.1148/ryai.2021210030.
• Centre for Medicare and Medicaid Services. Process and Information Required for a New Technology Ambulatory Payment Classification (APC) Assignment Under the Hospital Outpatient Prospective Payment System (OPPS). (2021).
• Kassing P, Berry CD. Hospital outpatient prospective payment system: a maturing prospective payment system. J. Am. Coll. Radiol. 2020;17:534–541. doi: 10.1016.